Neural Hermeneutics

     The term, hermeneutics, originally referred to the art of interpreting complex written texts, in particular holy scriptures, which demand considerable skill to reveal their meaning. Hermeneutics is especially relevant to the problem of translation, which, of necessity requires interpretation of the original text. The major problem for hermeneutics concerns how to develop criteria for deciding when an interpretation is correct.

     Modern practitioners of hermeneutics recognised that this problem applies, more generally, to any situation in which messages have to be interpreted. Schleiermacher, among others, suggested that the problem for the interpreter is to reveal the message intended by the speaker, i.e. what the author had in mind when she wrote the text (mens auctoris). However, this criterion, that the interpretation matches the intention, is problematic since the author might be dead, or otherwise unavailable. So all we typically have is the text and some knowledge of the context in which it was written.

     This difficulty is not confined to the interpretation of ancient texts. Even if I am talking with you face-to-face, I cannot access your mind to check whether my interpretation of what you have just said corresponds to what you intended me to understand. I can create a coherent story, but I can never get independent evidence about the correctness of my interpretations. Nevertheless, in spite of this apparently insurmountable difficulty, most of the time people seem to be able to understand each other very adequately. How is this achieved?

     Neural Hermeneutics is concerned with the mechanisms, instantiated in the brain, through which people are able to understand one another. Such mechanisms, although they might now be specialised for understanding, will have evolved from earlier mechanisms with other purposes. Two such extant mechanisms seem relevant to the problem of understanding. The first is predictive coding (or Bayesian inference) which explains our perception of the physical world. The second is simulation and alignment which aids our perception of the social world.

     We perceive an object, such as a tree, on the basis of signals from our senses. This process is not linear, since no sensation can unambiguously indicate the presence of a tree. Rather a computational loop is required circling from sensation to inference and back again. Our brain infers the most likely cause of the sensations and then tests this inference by collecting more sensory evidence (e.g. by moving the eyes, or touching the object). If the evidence is not what was expected on the basis of the inferred cause of the sensations (a prediction error) then the inference has to be updated. Once the fit between sensations and inferred cause is sufficiently good, the object is unambiguously perceived.

     In principle the same mechanism can be applied when trying to understand the mental world of others. The major difference is that, unlike with trees, the process goes in both directions: while I am trying to understand you, you are trying to understand me. Here the sensory evidence might be the words I hear from which I infer the idea you are trying to convey. I can test my inference, not only by predicting what else you are likely to say, but also by saying something myself and predicting how you will respond. Meanwhile you will be applying the same strategy to what I say. When our prediction errors become sufficiently low, then we have probably understood one another. In this account, the error we are minimising is not the difference between my idea and your idea, since we have no direct access to each other’s ideas. Rather, it is the difference between my idea and my representation of your idea .

     One advantage of the formulation in terms of predictive coding is that it elegantly captures the concept of the Hermeneutic Circle, whereby the whole cannot be understood without reference to the parts, while, at the same time, the parts cannot be understood without reference to the whole. In the same way, in the predictive coding loop, the inferred cause (the idea, the whole) predicts the evidence, while, at the same time, the evidence (the words, the parts) modifies the inferred cause.

     The predictive coding model outlined here does not explain how the link is made between the words and the idea, that is, how the initial inference is made. One possibility is to use simulation, that is I can predict what words you will use on the basis of what I myself would say in the same situation. This is by analogy with motor simulation, in which I predict the movements of others on the basis of my own motor system. There is considerable evidence for such a mechanism. A necessary consequence of the application of simulation to understanding others is that understanding will be more difficult to achieve if you are in some way different to the person you are trying to understand. This problem may be mitigated through alignment (or mirroring).

     We all have a strong and automatic tendency to imitate each other: the chameleon effect. This imitation or mirroring occurs in many domains, including gestures, emotions and aspects of speech such as intonation, grammar and vocabulary. Such mirroring makes us more similar to the person we are interacting with and thereby makes motor and mental simulation more efficient. Direct evidence that understanding is improved by alignment comes from a study showing that communication was improved when participants deliberately imitated the accent of the person they were talking to.

     The interactive mechanism I have described above implies that understanding is, in part, a collaboration between the partners engaged in the discourse. Thus I learn more about my own ideas through interacting with someone else. This relates to Schleiermacher’s suggestion that, by taking the context into account, the translator can achieve a better understanding of the text even than the original author. By the same argument a listener can have a better understanding of the speaker, than the speaker herself. This is because the listener will not only understand the message that the speaker intends to convey, but can also take account of signs, such as body language, indicating aspects of the message that the speaker was unaware of. This better understanding will be fed back to the speaker in the course of the conversation. Thus, through interactions with others we can achieve a better understanding of ourselves.

 

(thanks to Thomas Schwarz Wentzer)

Comments